The Network Connection
In order for the nodes in a cluster to communicate, they must talk with each other over the network. There are two different ways to enable this communication. First, they can share the same network connection (public network) as the users who access the cluster. Second, they can communicate over a private network (where the two nodes share a separate network).
Nodes in a cluster need to send and receive what is called a heartbeat signal, among other communications. This signal is used by each node to determine if the other node is still available. While this heartbeat signal can be configured to go over the public network, it is better if it is configured to run over a private network.
A private network is simply two network cards (one in each node of the cluster) that are connected via a pass-through network cable, or connected to each other by a hub. Personally, I prefer a direct connection using a pass-through cable because it eliminates another potential point of failure: the hub.
Using a private network produces four benefits. First, it moves the heartbeat traffic off the public network and onto the private network. This helps to reduce network congestion on the public network. Second, it helps to boost redundancy by eliminating a single point of failure. And third, it helps to boost network security. Last of all, if you don’t use a private network, Microsoft will not support your cluster (it is a requirement of the HCL).
For your public network connection, you may want to consider a 1GB or a 100MB network card, while a 100MB or a 10MB network card is adequate for the private network.
Using Identical Hardware
Although it is not absolutely required, ideally, the nodes in your cluster should have identical hardware, drivers, and configuration. In fact, the cards in each server should ideally be placed in the same numbered slot. For example, if you put a 100MB public network card in slot 2 of node 1, then you should also put the same 100MB network card in slot 2 of node 2.
Using identical hardware, drivers, and configuration produces many benefits, including:
- Less problems and troubleshooting. This is the main reason.
- Ensuring that both nodes have identical capacity to better handle production loads if failover occurs.
- Should you have major hardware problems and the worst happens, and can’t get parts fast enough, you can always cannibalize parts in one server to get the other one working.
Use the Correct, and Latest Hardware Drivers
Another big gotcha when configuring hardware is using outdated, buggy, or using the wrong driver for the node’s hardware. Many types of hardware have many different variations, each often requiring their own driver versions. It can often be very difficult to determine which driver you need, where to find it, and even how to properly install it. I have seen cases where parts and drivers have been mislabeled, which really can cause headaches.
So before you even begin build your servers, you will want to take time to research all of the drivers that you need, locate them, ensure that they are the latest versions, and also verify that they support Windows 2000 clustering. Just because a particular driver works after the server is build does not mean it will continue to work once Clustering Services has been installed.
If you are not careful, you could end up spending endless hours troubleshooting your cluster if you don’t have all the correct drivers installed correctly. And as with hardware, be sure both nodes in the cluster use the same drivers, configured identically.
Hardware Compatibility List
None of the advice given above will do you any good if the hardware you are using in your cluster is not certified by Microsoft, as an approved and tested cluster system. So what does this mean? Microsoft certifies hardware two different ways. First, individual parts can be certified to be Windows 2000 Cluster compatible. But that is not enough. Just because each of the components in your cluster have been certified individually does not mean your cluster hardware has a whole has been certified. Second, Microsoft certifies cluster hardware as total systems, and this is the certification you must get before you order the parts for your cluster.
To find out what “total systems” Microsoft certifies for Windows 2000 clustering, you must go to this url: www.microsoft.com/hcl/. From here, there are two options on the screen: “Search for the following” and “In the following types.” Under the option, “Search for the following,” select the option, “All products,” (the default), and for “In the following types, select “cluster” (not the default). Next, click on “Search now.”
The results of this return all of the approved hardware systems for Windows 2000 clustering. As you can see, there are a lot. But the system you want may not be there. If this is the case, then your system will not be supported by Microsoft if you run into problems. Because of this, and to reduce the chance for potential problems, you should only use approved hardware systems for your cluster.
So what exactly is a “hardware system?” The best way find the answer is to look at an example from the approved cluster systems approved by Microsoft for Windows 2000. For example, when you look up this system on this web page, “Compaq ProLiant Cluster HA F100/F200 (DL380)”, note that it is approved for Windows 2000 clustering. When you click on this link, a new window appears providing this information:
Compaq ProLiant Cluster HA F100/F200 (DL380)
Server 1: Compaq ProLiant DL380
Server 2: Compaq ProLiant DL380
SCSI/Raid Controller – Compaq StorageWorks Fibre Channel Host Adapter/P and Compaq StorageWorks 64-bit/66-Mhz Fibre Channel Host Adapter
Shared SCSI/RAID Storage – Compaq StorageWorks RAID Array 4000/4100
This information compromises the “hardware system.” It includes the servers acting as the nodes in the cluster, along with the tested and approved shared arrays to be used with the server nodes. So in other words, a cluster’s servers and its shared array must be approved as a system. On the other hand, other devices in your servers, such as network cards or graphics cards, are not considered part of the total hardware system. Instead, they must be also on the HCL for Windows 2000.
Where To Next?
As you can see from this article, selecting hardware for your Windows 2000 cluster is not an easy task. It involves much time and research to ensure you have all the right hardware, properly installed and connected.
Assuming you have done this part now, the next part, installing and configuring Windows 2000, should be relatively easy.
]]>